Whamcloud - gitweb
LU-9230 ldlm: speed up preparation for list of lock cancel
[fs/lustre-release.git] / lustre / ldlm / ldlm_lock.c
index 472dde5..41f4907 100644 (file)
@@ -23,7 +23,7 @@
  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2010, 2016, Intel Corporation.
+ * Copyright (c) 2010, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -236,6 +236,8 @@ int ldlm_lock_remove_from_lru_nolock(struct ldlm_lock *lock)
                struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
 
                LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
+               if (ns->ns_last_pos == &lock->l_lru)
+                       ns->ns_last_pos = lock->l_lru.prev;
                list_del_init(&lock->l_lru);
                LASSERT(ns->ns_nr_unused > 0);
                ns->ns_nr_unused--;
@@ -286,7 +288,6 @@ void ldlm_lock_add_to_lru_nolock(struct ldlm_lock *lock)
        LASSERT(list_empty(&lock->l_lru));
        LASSERT(lock->l_resource->lr_type != LDLM_FLOCK);
        list_add_tail(&lock->l_lru, &ns->ns_unused_list);
-       ldlm_clear_skipped(lock);
        LASSERT(ns->ns_nr_unused >= 0);
        ns->ns_nr_unused++;
 }
@@ -868,7 +869,8 @@ void ldlm_lock_decref_internal(struct ldlm_lock *lock, enum ldlm_mode mode)
         } else if (ns_is_client(ns) &&
                    !lock->l_readers && !lock->l_writers &&
                   !ldlm_is_no_lru(lock) &&
-                  !ldlm_is_bl_ast(lock)) {
+                  !ldlm_is_bl_ast(lock) &&
+                  !ldlm_is_converting(lock)) {
 
                 LDLM_DEBUG(lock, "add lock into lru list");
 
@@ -1072,16 +1074,14 @@ static void ldlm_granted_list_add_lock(struct ldlm_lock *lock,
  * Add a lock to granted list on a resource maintaining skiplist
  * correctness.
  */
-static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
+void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
 {
-        struct sl_insert_point prev;
-        ENTRY;
+       struct sl_insert_point prev;
 
-        LASSERT(lock->l_req_mode == lock->l_granted_mode);
+       LASSERT(lock->l_req_mode == lock->l_granted_mode);
 
-        search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
-        ldlm_granted_list_add_lock(lock, &prev);
-        EXIT;
+       search_granted_lock(&lock->l_resource->lr_granted, lock, &prev);
+       ldlm_granted_list_add_lock(lock, &prev);
 }
 
 /**
@@ -1091,7 +1091,6 @@ static void ldlm_grant_lock_with_skiplist(struct ldlm_lock *lock)
  * NOTE: called by
  *  - ldlm_lock_enqueue
  *  - ldlm_reprocess_queue
- *  - ldlm_lock_convert
  *
  * must be called with lr_lock held
  */
@@ -1414,9 +1413,6 @@ enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
                GOTO(out, rc = 1);
        if (flags & LDLM_FL_BLOCK_GRANTED)
                GOTO(out, rc = 0);
-       lock = search_queue(&res->lr_converting, &data);
-       if (lock != NULL)
-               GOTO(out, rc = 1);
        lock = search_queue(&res->lr_waiting, &data);
        if (lock != NULL)
                GOTO(out, rc = 1);
@@ -1695,6 +1691,33 @@ out:
        RETURN(ERR_PTR(rc));
 }
 
+#ifdef HAVE_SERVER_SUPPORT
+static enum ldlm_error ldlm_lock_enqueue_helper(struct ldlm_lock *lock,
+                                            __u64 *flags)
+{
+       struct ldlm_resource *res = lock->l_resource;
+       enum ldlm_error rc = ELDLM_OK;
+       struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
+       ldlm_processing_policy policy;
+       ENTRY;
+
+       policy = ldlm_processing_policy_table[res->lr_type];
+restart:
+       policy(lock, flags, LDLM_PROCESS_ENQUEUE, &rc, &rpc_list);
+       if (rc == ELDLM_OK && lock->l_granted_mode != lock->l_req_mode &&
+           res->lr_type != LDLM_FLOCK) {
+               rc = ldlm_handle_conflict_lock(lock, flags, &rpc_list);
+               if (rc == -ERESTART)
+                       GOTO(restart, rc);
+       }
+
+       if (!list_empty(&rpc_list))
+               ldlm_discard_bl_list(&rpc_list);
+
+       RETURN(rc);
+}
+#endif
+
 /**
  * Enqueue (request) a lock.
  *
@@ -1712,9 +1735,6 @@ enum ldlm_error ldlm_lock_enqueue(struct ldlm_namespace *ns,
        struct ldlm_lock *lock = *lockp;
        struct ldlm_resource *res = lock->l_resource;
        int local = ns_is_client(ldlm_res_to_ns(res));
-#ifdef HAVE_SERVER_SUPPORT
-       ldlm_processing_policy policy;
-#endif
        enum ldlm_error rc = ELDLM_OK;
        struct ldlm_interval *node = NULL;
        ENTRY;
@@ -1814,33 +1834,27 @@ enum ldlm_error ldlm_lock_enqueue(struct ldlm_namespace *ns,
         * more or less trusting the clients not to lie.
         *
         * FIXME (bug 268): Detect obvious lies by checking compatibility in
-        * granted/converting queues. */
+        * granted queue. */
         if (local) {
-                if (*flags & LDLM_FL_BLOCK_CONV)
-                        ldlm_resource_add_lock(res, &res->lr_converting, lock);
-                else if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
-                        ldlm_resource_add_lock(res, &res->lr_waiting, lock);
-                else
-                        ldlm_grant_lock(lock, NULL);
+               if (*flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED))
+                       ldlm_resource_add_lock(res, &res->lr_waiting, lock);
+               else
+                       ldlm_grant_lock(lock, NULL);
                GOTO(out, rc = ELDLM_OK);
 #ifdef HAVE_SERVER_SUPPORT
-        } else if (*flags & LDLM_FL_REPLAY) {
-                if (*flags & LDLM_FL_BLOCK_CONV) {
-                        ldlm_resource_add_lock(res, &res->lr_converting, lock);
-                       GOTO(out, rc = ELDLM_OK);
-                } else if (*flags & LDLM_FL_BLOCK_WAIT) {
-                        ldlm_resource_add_lock(res, &res->lr_waiting, lock);
+       } else if (*flags & LDLM_FL_REPLAY) {
+               if (*flags & LDLM_FL_BLOCK_WAIT) {
+                       ldlm_resource_add_lock(res, &res->lr_waiting, lock);
                        GOTO(out, rc = ELDLM_OK);
-                } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
-                        ldlm_grant_lock(lock, NULL);
+               } else if (*flags & LDLM_FL_BLOCK_GRANTED) {
+                       ldlm_grant_lock(lock, NULL);
                        GOTO(out, rc = ELDLM_OK);
-                }
-                /* If no flags, fall through to normal enqueue path. */
-        }
+               }
+               /* If no flags, fall through to normal enqueue path. */
+       }
 
-        policy = ldlm_processing_policy_table[res->lr_type];
-       policy(lock, flags, LDLM_PROCESS_ENQUEUE, &rc, NULL);
-        GOTO(out, rc);
+       rc = ldlm_lock_enqueue_helper(lock, flags);
+       GOTO(out, rc);
 #else
         } else {
                 CERROR("This is client-side-only module, cannot handle "
@@ -1872,6 +1886,7 @@ int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
        __u64 flags;
        int rc = LDLM_ITER_CONTINUE;
        enum ldlm_error err;
+       struct list_head bl_ast_list = LIST_HEAD_INIT(bl_ast_list);
        ENTRY;
 
        check_res_locked(res);
@@ -1881,15 +1896,23 @@ int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
        LASSERT(intention == LDLM_PROCESS_RESCAN ||
                intention == LDLM_PROCESS_RECOVERY);
 
+restart:
        list_for_each_safe(tmp, pos, queue) {
                struct ldlm_lock *pending;
+               struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
 
                pending = list_entry(tmp, struct ldlm_lock, l_res_link);
 
                 CDEBUG(D_INFO, "Reprocessing lock %p\n", pending);
 
                 flags = 0;
-               rc = policy(pending, &flags, intention, &err, work_list);
+               rc = policy(pending, &flags, intention, &err, &rpc_list);
+               if (pending->l_granted_mode == pending->l_req_mode ||
+                   res->lr_type == LDLM_FLOCK) {
+                       list_splice(&rpc_list, work_list);
+               } else {
+                       list_splice(&rpc_list, &bl_ast_list);
+               }
                /*
                 * When this is called from recovery done, we always want
                 * to scan the whole list no matter what 'rc' is returned.
@@ -1899,6 +1922,20 @@ int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
                        break;
         }
 
+       if (!list_empty(&bl_ast_list)) {
+               unlock_res(res);
+
+               rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &bl_ast_list,
+                                      LDLM_WORK_BL_AST);
+
+               lock_res(res);
+               if (rc == -ERESTART)
+                       GOTO(restart, rc);
+       }
+
+       if (!list_empty(&bl_ast_list))
+               ldlm_discard_bl_list(&bl_ast_list);
+
         RETURN(intention == LDLM_PROCESS_RESCAN ? rc : LDLM_ITER_CONTINUE);
 }
 
@@ -1909,7 +1946,6 @@ int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
  * \param[in] lock             The lock to be enqueued.
  * \param[out] flags           Lock flags for the lock to be enqueued.
  * \param[in] rpc_list         Conflicting locks list.
- * \param[in] grant_flags      extra flags when granting a lock.
  *
  * \retval -ERESTART:  Some lock was instantly canceled while sending
  *                     blocking ASTs, caller needs to re-check conflicting
@@ -1918,7 +1954,7 @@ int ldlm_reprocess_queue(struct ldlm_resource *res, struct list_head *queue,
  * \reval 0:           Lock is successfully added in waiting list.
  */
 int ldlm_handle_conflict_lock(struct ldlm_lock *lock, __u64 *flags,
-                             struct list_head *rpc_list, __u64 grant_flags)
+                             struct list_head *rpc_list)
 {
        struct ldlm_resource *res = lock->l_resource;
        int rc;
@@ -1967,7 +2003,7 @@ int ldlm_handle_conflict_lock(struct ldlm_lock *lock, __u64 *flags,
 
                RETURN(rc);
        }
-       *flags |= (LDLM_FL_BLOCK_GRANTED | grant_flags);
+       *flags |= LDLM_FL_BLOCK_GRANTED;
 
        RETURN(0);
 }
@@ -2029,6 +2065,13 @@ ldlm_work_bl_ast_lock(struct ptlrpc_request_set *rqset, void *opaq)
        unlock_res_and_lock(lock);
 
        ldlm_lock2desc(lock->l_blocking_lock, &d);
+       /* copy blocking lock ibits in cancel_bits as well,
+        * new client may use them for lock convert and it is
+        * important to use new field to convert locks from
+        * new servers only
+        */
+       d.l_policy_data.l_inodebits.cancel_bits =
+               lock->l_blocking_lock->l_policy_data.l_inodebits.bits;
 
        rc = lock->l_blocking_ast(lock, &d, (void *)arg, LDLM_CB_BLOCKING);
        LDLM_LOCK_RELEASE(lock->l_blocking_lock);
@@ -2216,7 +2259,7 @@ out:
 /**
  * Try to grant all waiting locks on a resource.
  *
- * Calls ldlm_reprocess_queue on converting and waiting queues.
+ * Calls ldlm_reprocess_queue on waiting queue.
  *
  * Typically called after some resource locks are cancelled to see
  * if anything could be granted as a result of the cancellation.
@@ -2246,19 +2289,15 @@ static void __ldlm_reprocess_all(struct ldlm_resource *res,
                RETURN_EXIT;
 restart:
        lock_res(res);
-       rc = ldlm_reprocess_queue(res, &res->lr_converting, &rpc_list,
-                                 intention);
-       if (rc == LDLM_ITER_CONTINUE)
-               ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list,
-                                    intention);
+       ldlm_reprocess_queue(res, &res->lr_waiting, &rpc_list, intention);
        unlock_res(res);
 
-        rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
-                               LDLM_WORK_CP_AST);
-        if (rc == -ERESTART) {
+       rc = ldlm_run_ast_work(ldlm_res_to_ns(res), &rpc_list,
+                              LDLM_WORK_CP_AST);
+       if (rc == -ERESTART) {
                LASSERT(list_empty(&rpc_list));
-                goto restart;
-        }
+               goto restart;
+       }
 #else
         ENTRY;
 
@@ -2303,19 +2342,6 @@ void ldlm_reprocess_recovery_done(struct ldlm_namespace *ns)
        EXIT;
 }
 
-static bool is_bl_done(struct ldlm_lock *lock)
-{
-       bool bl_done = true;
-
-       if (!ldlm_is_bl_done(lock)) {
-               lock_res_and_lock(lock);
-               bl_done = ldlm_is_bl_done(lock);
-               unlock_res_and_lock(lock);
-       }
-
-       return bl_done;
-}
-
 /**
  * Helper function to call blocking AST for LDLM lock \a lock in a
  * "cancelling" mode.
@@ -2379,6 +2405,7 @@ void ldlm_lock_cancel(struct ldlm_lock *lock)
          * talking to me first. -phik */
         if (lock->l_readers || lock->l_writers) {
                 LDLM_ERROR(lock, "lock still has references");
+               unlock_res_and_lock(lock);
                 LBUG();
         }
 
@@ -2441,7 +2468,7 @@ static void ldlm_cancel_lock_for_export(struct obd_export *exp,
 
        res = ldlm_resource_getref(lock->l_resource);
 
-       ldlm_res_lvbo_update(res, NULL, 1);
+       ldlm_lvbo_update(res, lock, NULL, 1);
        ldlm_lock_cancel(lock);
        if (!exp->exp_obd->obd_stopping)
                ldlm_reprocess_all(res);
@@ -2542,18 +2569,18 @@ int ldlm_export_cancel_locks(struct obd_export *exp)
 }
 
 /**
- * Downgrade an exclusive lock.
+ * Downgrade an PW/EX lock to COS mode.
  *
- * A fast variant of ldlm_lock_convert for convertion of exclusive locks. The
+ * A lock mode convertion from PW/EX mode to less conflict mode. The
  * convertion may fail if lock was canceled before downgrade, but it doesn't
  * indicate any problem, because such lock has no reader or writer, and will
  * be released soon.
- * Used by Commit on Sharing (COS) code.
+ * Used by Commit on Sharing (COS) code only for now.
  *
  * \param lock A lock to convert
  * \param new_mode new lock mode
  */
-void ldlm_lock_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode)
+void ldlm_lock_mode_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode)
 {
        ENTRY;
 
@@ -2584,137 +2611,7 @@ void ldlm_lock_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode)
 
        EXIT;
 }
-EXPORT_SYMBOL(ldlm_lock_downgrade);
-
-/**
- * Attempt to convert already granted lock to a different mode.
- *
- * While lock conversion is not currently used, future client-side
- * optimizations could take advantage of it to avoid discarding cached
- * pages on a file.
- */
-struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock,
-                                       enum ldlm_mode new_mode, __u32 *flags)
-{
-       struct list_head rpc_list;
-       struct ldlm_resource *res;
-       struct ldlm_namespace *ns;
-       int granted = 0;
-#ifdef HAVE_SERVER_SUPPORT
-       int old_mode;
-       struct sl_insert_point prev;
-#endif
-       struct ldlm_interval *node;
-       ENTRY;
-
-       INIT_LIST_HEAD(&rpc_list);
-       /* Just return if mode is unchanged. */
-       if (new_mode == lock->l_granted_mode) {
-               *flags |= LDLM_FL_BLOCK_GRANTED;
-               RETURN(lock->l_resource);
-       }
-
-       /* I can't check the type of lock here because the bitlock of lock
-        * is not held here, so do the allocation blindly. -jay */
-       OBD_SLAB_ALLOC_PTR_GFP(node, ldlm_interval_slab, GFP_NOFS);
-       if (node == NULL)  /* Actually, this causes EDEADLOCK to be returned */
-               RETURN(NULL);
-
-       LASSERTF((new_mode == LCK_PW && lock->l_granted_mode == LCK_PR),
-                "new_mode %u, granted %u\n", new_mode, lock->l_granted_mode);
-
-       lock_res_and_lock(lock);
-
-       res = lock->l_resource;
-       ns  = ldlm_res_to_ns(res);
-
-#ifdef HAVE_SERVER_SUPPORT
-       old_mode = lock->l_req_mode;
-#endif
-       lock->l_req_mode = new_mode;
-       if (res->lr_type == LDLM_PLAIN || res->lr_type == LDLM_IBITS) {
-#ifdef HAVE_SERVER_SUPPORT
-               /* remember the lock position where the lock might be
-                * added back to the granted list later and also
-                * remember the join mode for skiplist fixing. */
-               prev.res_link = lock->l_res_link.prev;
-               prev.mode_link = lock->l_sl_mode.prev;
-               prev.policy_link = lock->l_sl_policy.prev;
-#endif
-                ldlm_resource_unlink_lock(lock);
-        } else {
-                ldlm_resource_unlink_lock(lock);
-                if (res->lr_type == LDLM_EXTENT) {
-                        /* FIXME: ugly code, I have to attach the lock to a
-                         * interval node again since perhaps it will be granted
-                         * soon */
-                       INIT_LIST_HEAD(&node->li_group);
-                        ldlm_interval_attach(node, lock);
-                        node = NULL;
-                }
-        }
-
-        /*
-         * Remove old lock from the pool before adding the lock with new
-         * mode below in ->policy()
-         */
-        ldlm_pool_del(&ns->ns_pool, lock);
-
-        /* If this is a local resource, put it on the appropriate list. */
-        if (ns_is_client(ldlm_res_to_ns(res))) {
-                if (*flags & (LDLM_FL_BLOCK_CONV | LDLM_FL_BLOCK_GRANTED)) {
-                        ldlm_resource_add_lock(res, &res->lr_converting, lock);
-                } else {
-                        /* This should never happen, because of the way the
-                         * server handles conversions. */
-                       LDLM_ERROR(lock, "Erroneous flags %x on local lock\n",
-                                   *flags);
-                        LBUG();
-
-                        ldlm_grant_lock(lock, &rpc_list);
-                        granted = 1;
-                        /* FIXME: completion handling not with lr_lock held ! */
-                        if (lock->l_completion_ast)
-                                lock->l_completion_ast(lock, 0, NULL);
-                }
-#ifdef HAVE_SERVER_SUPPORT
-       } else {
-               int rc;
-               enum ldlm_error err;
-               __u64 pflags = 0;
-               ldlm_processing_policy policy;
-
-                policy = ldlm_processing_policy_table[res->lr_type];
-               rc = policy(lock, &pflags, LDLM_PROCESS_RESCAN, &err,
-                           &rpc_list);
-                if (rc == LDLM_ITER_STOP) {
-                        lock->l_req_mode = old_mode;
-                        if (res->lr_type == LDLM_EXTENT)
-                                ldlm_extent_add_lock(res, lock);
-                        else
-                                ldlm_granted_list_add_lock(lock, &prev);
-
-                        res = NULL;
-                } else {
-                        *flags |= LDLM_FL_BLOCK_GRANTED;
-                        granted = 1;
-                }
-        }
-#else
-        } else {
-                CERROR("This is client-side-only module, cannot handle "
-                       "LDLM_NAMESPACE_SERVER resource type lock.\n");
-                LBUG();
-        }
-#endif
-        unlock_res_and_lock(lock);
-
-        if (granted)
-                ldlm_run_ast_work(ns, &rpc_list, LDLM_WORK_CP_AST);
-        if (node)
-                OBD_SLAB_FREE(node, ldlm_interval_slab, sizeof(*node));
-        RETURN(res);
-}
+EXPORT_SYMBOL(ldlm_lock_mode_downgrade);
 
 /**
  * Print lock with lock handle \a lockh description into debug log.
@@ -2764,10 +2661,10 @@ void _ldlm_lock_debug(struct ldlm_lock *lock,
         va_start(args, fmt);
 
         if (exp && exp->exp_connection) {
-                nid = libcfs_nid2str(exp->exp_connection->c_peer.nid);
+               nid = obd_export_nid2str(exp);
         } else if (exp && exp->exp_obd != NULL) {
                 struct obd_import *imp = exp->exp_obd->u.cli.cl_import;
-                nid = libcfs_nid2str(imp->imp_connection->c_peer.nid);
+               nid = obd_import_nid2str(imp);
         }
 
         if (resource == NULL) {